TradingView ICT Master Suite Best Settings

How we searched ICT Master Suite’s settings

In September we backtested ICT Master Suite [Trading IQ], an open-source TradingView strategy script, on Nasdaq futures. Its best default setup made +17.36% in fourteen weeks, and all of that profit came from 15 trades in one session. Ported to Python and run long-only over 8.5 years at the script’s own default cost setting, it lost money with no profitable year. That article tested the script’s ten default setups. This one searches its other settings, on the full run of Nasdaq futures data we could gather, 2016 through early September 2026.

We searched 398 combinations of the script’s settings across the three models we ported (Silver Bullet, Model 2022, Liquidity Raid): the timeframe, the liquidity-sweep switches and sliders, the take-profit and stop-loss choices, and, for every model, a 15-minute bias filter and a trading-hour window. Only Silver Bullet has those last two in the script, so for Model 2022 and Liquidity Raid we added them in our own code, to compare the models on equal footing. Two models in the script, Unicorn and OTE, were not ported, so this says nothing about their settings.

We picked the single best combination on 2016–2022 by its per-trade Sharpe ratio (return per unit of risk), among combinations with at least 100 trades and a result unlikely to be chance. That is one reasonable rule among others we did not try. Then we reported the winner once, unchanged, on 2023 through September 2026. It is the setting in the box at the top: Model 2022, long only, on 5 minutes, with both added filters on.

The first three rows use the same cost and fill assumptions as throughout; the last row reprices the identical trades at the script’s own default commission (0.02% of the entry price per side) instead, shown here for comparison. The later-period window was fixed before any setting was chosen; the setting itself was fixed before the later period was looked at.

The best setting kept making money on the later period, in every year of it, from +0.3R in 2024 to +15.2R in 2023. But per trade it did no better than the same trades placed on random dates, and once we account for having tried 398 different settings, the same kind of correction applied to every other claim in this article, none of the 398 clears the bar, including the single best one. At the script’s own default commission, the same trades lose clearly.

One check points the other way. Trading one NQ contract per trade, instead of risking the same amount each time, the best setting made +$126,290 on 2023 – Sept. 2026, more than every one of 500 random-date versions of the same trades, each sized at the stop of the trade it replays. Two things limit what that means. Most of the profit came from about ten large winners, and the trades with wider stops did far better per unit of risk (+0.16R on average) than the tighter half (+0.01R), which one-contract sizing rewards. And we added this check after the planned one had come back unremarkable, so it has none of the planned test’s protection against finding a pattern by looking for one. It is a lead to test again on new data, not evidence of an edge.

Switching the two added filters off leaves only settings the script really has. That version traded about two and a half times as often and won three trades in four, but each win averaged about a third the size of each loss. Its profit factor was 1.05 on 2016–2022 (1.00 with a tick of slippage) and 0.98 on 2023–2026. The script’s defaults, Model 2022 long and short, came in at 1.01 and then 0.87.

How closely each port was checked against the real script differs. Silver Bullet’s trade-by-trade logic was checked earlier against a real TradingView trade list (43 of 61 trades landed on the same bar, 40 of those 43 agreed on win or loss); this study’s additions on top of it — the sweep-gate overlay, the wider settings grid, the short side — were checked for internal consistency rather than against a fresh trade list. Model 2022 was checked by reading the Pine code directly and by a self-consistency test, not against a trade list of its own, since none exists for it; its own rule for spotting a liquidity sweep is a documented simplification of the published script’s more elaborate version. Liquidity Raid’s entry condition, a “breaker block,” uses a simpler stand-in for the published script’s own order-block bookkeeping, checked for correctness in the pipeline rather than hand-checked trade by trade against the Pine source — one more reason it carries no weight on its own in the six tweak tests below.

A bug from the last Master Suite study, still there

The earlier ICT Master Suite article found that the script’s shared library reads the long side’s stop-loss setting on the short side’s own code path, so Silver Bullet Short’s Silver Bullet Short Strategy SL setting has no effect; its stop always falls back to the day’s high, regardless of what a trader sets it to. Checked directly against the current library source (version 6.0), that bug is still there.

We measured how much it matters, comparing results as the bug actually behaves against results if the short side’s own setting were respected, across all 72 paired settings, measured on the 2016–2022 portion of the data. The comparison shows a consistent pattern: the bug’s day-high fallback stop produced a higher average result than the trader’s own chosen stop setting would have, in 53 of the 72 paired settings, for an average difference of about three hundredths of a risk unit per trade — not large, but a direction unlikely to be chance.

Six tweaks, tested one at a time

Alongside the settings search, we tested six tweaks to these setups, one at a time: require the liquidity sweep; take direction from the sweep; enter on the 1-minute fair value gap rather than the 5-minute one; trade every qualifying setup, often; drop the 15-minute directional bias; and don’t restrict trading to the famous kill-zone hours.

We mapped each tweak onto ICT Master Suite and tested it on Nasdaq futures minute bars. Where the script has a real setting for a tweak, it’s named below exactly as the TradingView input labels it; where it doesn’t, we built our own clearly labelled stand-in and say so each time.

Before looking at results, we fixed which single number would decide each tweak, and applied a stricter bar for having tested several ideas at once. Silver Bullet and Model 2022 carry most of the weight below; Liquidity Raid is part of the settings search above but did not produce a usable result for any of the six tweaks on its own.

1. Requiring the liquidity sweep

The tweak: only trade fair value gaps that form after a liquidity sweep. The real setting it maps to is Silver Bullet’s Only FVGs Formed After Liquidity Sweep, off by default, together with Liquidity Sweep Aggressiveness, a slider that controls how large a move has to be before it counts as a sweep.

Checked directly against the published Pine code, that real switch does nothing at all to Silver Bullet’s own trades: its entry logic never reads it or any fair-value-gap array keyed to a sweep. To test the idea on Silver Bullet anyway, we built our own on/off overlay that mimics what the switch is supposed to do, and toggled that overlay at three slider settings across NQ futures from 2016 to 2026. It produced no reliable difference in either direction — all six versions of that test, three slider settings, both long and short, came back inconclusive, with no real difference turning up either way. That result describes our own stand-in, not the real setting, which simply has no effect to measure.

Model 2022 is a different story, but a complicated one. In the script, Model 2022 forces the sweep requirement on; there is no real setting that turns it off. To test the idea on Model 2022 anyway, we built a stand-in: it kept Model 2022’s entry rule the same and swapped in a looser pivot detector for its stop and target, standing in for “no sweep requirement.” Requiring the sweep did better, not worse, by about two hundredths of a risk unit per trade than skipping it, pooled across Silver Bullet and Model 2022, long and short combined. That is the direction the tweak predicts, and it is unlikely to be chance: reshuffling the data 1,000 times never once produced a gap this large (that is the limit of how finely this test measures “unlikely,” not a precise odds). It held up at every cost level we tried and on a second, independent Nasdaq-100 feed too.

The catch is the stand-in’s stops. In a spot-check of the first half of 2016 they sat roughly 2.6 times closer to entry than the real gate’s stops. Results here are measured in units of risk, and the stand-in loses a little on average, so a tighter stop makes that small loss look bigger per unit of risk. That widens the gap in favor of requiring the sweep, and trading costs widen it further for the same reason. So part of the difference comes from how the stand-in was built, and this comparison is not a clean test of a real, selectable setting.

Every trade above: no restriction to a trading-hour window, 15-minute bias off, Liquidity Sweep Aggressiveness at 5 (and the matching Model 2022 slider at 5); when a bar’s path could have hit either the stop or the target, the stop counts; orders fill only once price actually reaches the level; no added cost; NQ futures minute bars 2016–2026. Both rows use this study’s own research overlays, not a real TradingView input: the top row is our stand-in for Silver Bullet’s real switch (which does nothing to its trades), and the bottom row is our stand-in for a Model 2022 “off” state that doesn’t exist as a setting. The difference grows, rather than shrinks, once a realistic per-trade cost is added — a side effect of the Model 2022 stand-in’s tighter stops, not independent confirmation.

2. Taking direction from the sweep

The tweak: take the trade’s direction from the sweep. This isn’t a separate toggle in the script; it’s the direction Silver Bullet and Model 2022 already trade in, by design: after a sweep below a level, look for a long from the fair value gap that follows. To test whether that direction is actually doing any work, we compared it against two controls that share everything else: a version where each trade’s direction label is reshuffled, and a mirror trade taken in the opposite direction at the identical stop and target distance.

Pooled across Silver Bullet and Model 2022 (about a quarter-point of cost per round turn, stop counted first on an ambiguous bar, orders filled only once price actually touched the level), real trades beat 1,000 reshuffled-direction versions of themselves on every single draw — a result very unlikely to be chance under reshuffling. But almost nine in ten of the pooled trades come from Model 2022, whose own stop and target are built from the sweep leg itself, which makes a positive result close to built-in rather than independent confirmation that “fade the sweep” travels as an idea.

Silver Bullet on its own does not confirm the idea cleanly. Its long side’s own mirror comparison favors taking direction from the sweep, by about 0.07R, though that range just touches zero; its short side’s own mirror comparison points the other way, by about −0.03R, a range that also touches zero. Judged against the reshuffled-direction control instead, the two sides disagree again: the short side beats its reshuffled versions, while the long side does clearly worse than its own, which goes against the idea.

Checked cold, with every setting copied unchanged, on a Nasdaq-100 CFD from a different data provider (Dukascopy), this idea does not hold up at all — not even the Model 2022 half, which had looked like the cleanest result in this whole study on NQ futures alone. The pooled result, which beat every one of a thousand reshuffles on NQ futures, lands in the bottom tenth of its own shuffled-direction range on Dukascopy. Model 2022’s own mirror-matched advantage, worth about four hundredths of a risk unit per trade in the tweak’s favor on NQ futures, flips to a small loss on Dukascopy. Silver Bullet short’s edge over the reshuffled-direction control, the one result on NQ futures that did favor the idea, also lands well inside its own null range on Dukascopy — not significant there either.

3. Entering on the 1-minute gap

The tweak: enter on the fair value gap on 1-minute bars instead of 5-minute ones. It maps directly onto the script’s own chart Timeframe input, which decides whether fair value gaps are read on 1-minute or 5-minute bars. For each of four model-and-direction combinations (Silver Bullet and Model 2022, long and short), we measured each timeframe’s own edge over a matched random-entry control, repeated 200 times with different random draws, then compared the 1-minute edge against the 5-minute edge, pooled across all four.

It came back the opposite way from the tweak, if only slightly. Pooled across Silver Bullet and Model 2022, 1-minute entries’ own edge over their matched random-timing control (+0.013R) was a touch smaller than 5-minute entries’ own edge (+0.016R) — a difference unlikely to be chance on its own, but one that does not survive a stricter bar once several ideas are tested together. The fairest summary: this test found no support for entering on the 1-minute gap over the 5-minute one.

4. Trading every setup, and trading often

The tweak: take every qualifying setup and trade often, since the edge per trade is small. We checked that idea across 2,700 different combinations of Silver Bullet’s settings, comparing how often each combination traded per year against how good its average trade was. We found no reliable relationship in either direction: this test did not show settings that traded far more often doing detectably better or worse, per trade, than settings that traded rarely.

A second check looked for a relationship between trading frequency and total profit instead of the average trade, and at first it looked strongly real. It wasn’t: total profit is just the average trade multiplied by the number of trades, so settings that traded more were always going to show bigger totals even if every trade carried the exact same edge. Once that arithmetic was accounted for with a matching comparison built to reproduce it, the observed relationship sat well within the range that arithmetic alone produces. No reliable frequency-to-profit relationship, in either direction, survived that correction.

Most of those 2,700 combinations are the same underlying trades, just re-sliced by cost, fee, and tie-rule choices that don’t change which trades happened — so this “how unlikely by chance” reading treats some closely related slices as if they were fully separate tests, which overstates how much evidence 2,700 combinations really provides. Repeating the comparison on only the 225 genuinely distinct underlying trade lists, with no re-slicing at all, gives the same answer: no reliable relationship either way.

5. Dropping the 15-minute bias

The tweak: switch the 15-minute directional bias off. Silver Bullet ships with exactly this bias: a 15-minute zigzag reading of the day’s direction that, when turned on, restricts trades to that side. Testing it on, versus off, inside Silver Bullet’s own three named kill zones, the bias made the average trade about two and a half hundredths of a risk unit less negative, not more — the opposite of what the tweak predicts. That difference clears an ordinary pass/fail line on its own, but once we account for having tested the bias across 25 different trading-hour windows, plus a few related checks run alongside them, it no longer clears a stricter one — one single hour among those windows does clear that stricter bar, but it isn’t one of Silver Bullet’s own kill zones.

Both rows: Silver Bullet’s native 15-minute bias, sweep requirement on, an ambiguous bar counts as a stop, orders fill only once price touches the level, roughly a quarter-point of cost per round turn, NQ futures 2016–2026.

A look-ahead bug in our own shared code (the bias was reading the bar’s open instead of its close, handing it a few minutes of foresight) was found and fixed mid-study; every number above is from the corrected version.

6. Were the famous trading windows special?

The tweak: don’t limit trading to the famous kill-zone hours, on the view that they aren’t special. Silver Bullet trades only inside three named one-hour windows, all New York time: 3–4 a.m. (overlapping the London session), 10–11 a.m. and 2–3 p.m. To check whether those specific hours are special, we ran the same sweep-and-fair-value-gap logic as a single-hour strategy, separately, on every hour of the day with enough data to trust (23 of 24 hours; one was dropped for thin coverage), and ranked them by average result.

On this test, the 10 a.m. hour, the middle of Silver Bullet’s three windows, ranked 4th best of 23 hours, with an average result just below breakeven and a win rate of 64.2% (range 62.6% to 65.8%). The rank is not pinned down tightly: resampling the underlying trading days puts the plausible range anywhere from 1st to 16th of 23. So “somewhere in the upper half, not clearly the single best” is a fair summary, and chance alone cannot be ruled out as the explanation for this specific rank.

So which settings work?

On this test, no ICT Master Suite setting held up by the bar we set in advance. The single best of 398 made money on Nasdaq futures in both periods, with a profit factor of 1.53 on the years it was picked from and 1.27 on the later years it never saw. But it needs two filters the script doesn’t have, its result per trade was no better than random dates, none of the 398 settings survives a correction for how many were tried, and at the script’s own default commission it loses. With only the script’s own inputs, the same setup about broke even and then lost, and the defaults lost on recent data. The one result that favours the best setting, its one-contract dollar total beating every random-date version, comes from a check we added afterwards and rests on a handful of large trades.

The six tweaks fared little better. Two held up against the stricter bar, requiring the sweep and taking direction from it, both in the direction the tweaks predict, but neither cleanly. The sweep requirement only works through an invented stand-in setting for Model 2022, which has no real on/off switch for this rule. Direction-from-the-sweep is carried almost entirely by that same model, whose own exits are built from the sweep leg, and when we checked it, with every setting unchanged, on a second Nasdaq-100 feed, it did not hold up there, Model 2022 included. The other four tweaks, entering on the 1-minute gap, dropping the 15-minute bias, treating the famous hours as unremarkable and trading often for its own sake, either pointed the opposite way before correction or showed no reliable relationship at all.

That is not quite the same as saying nothing here is real. Model 2022’s sweep-requirement and direction-from-the-sweep results are real patterns in the Nasdaq futures data, stable across a range of costs. But both are close to built into that one model’s own construction rather than independent evidence that the ideas work on their own, and the direction result does not repeat on a second data feed.

Nothing here is a reason to trade any of these settings, with or without the two added filters. A trader’s own discretionary judgment — which of several candidate setups a trader would actually take on a given morning — was not and could not be tested here; every number above is our own mechanical approximation of the Pine code, built to use only information available at the time of each trade.

Questions people ask